# Copyright 2025 The EasyDeL Author @erfanzar (Erfan Zare Chavoshi).
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import math
from jax.sharding import PartitionSpec
from easydel.infra.base_module import EasyDeLBaseConfig
from easydel.infra.etils import EasyDeLGradientCheckPointers
from easydel.infra.factory import register_config
[docs]@register_config("mamba2")
class Mamba2Config(EasyDeLBaseConfig):
"""
Configuration objects inherit from [`EasyDeLBaseConfig`] and can be used to control the model outputs. Read
the documentation from [`EasyDeLBaseConfig`] for more information.
Args:
vocab_size (`int`, *optional*, defaults to 32768):
Vocabulary size of the Mamba2 model. Defines the number of different tokens that can be represented by the
`inputs_ids` passed to the forward method.
hidden_size (`int`, *optional*, defaults to 4096):
Dimensionality of the encoder layers and the pooler layer.
state_size (`int`, *optional*, defaults to 128):
State size of the Mamba2 model.
num_hidden_layers (`int`, *optional*, defaults to 64):
Number of hidden layers in the Mamba2 encoder.
num_heads (`int`, *optional*, defaults to 128):
Number of attention heads for the grouped selective scan.
head_dim (`int`, *optional*, defaults to 64):
Dimension of each attention head.
n_groups (`int`, *optional*, defaults to 8):
Number of groups for the grouped selective scan.
layer_norm_epsilon (`float`, *optional*, defaults to 1e-5):
The epsilon used by the layer normalization layers.
pad_token_id (`int`, *optional*, defaults to 1):
The index of the padding token in the vocabulary.
bos_token_id (`int`, *optional*, defaults to 0):
The id of the *beginning-of-sequence* token.
eos_token_id (`int`, *optional*, defaults to 2):
The id of the *end-of-sequence* token.
expand (`int`, *optional*, defaults to 2):
Expansion factor for the intermediate size.
conv_kernel (`int`, *optional*, defaults to 4):
Kernel size of the convolution layer.
use_bias (`bool`, *optional*, defaults to `False`):
Whether to use bias in the linear layers.
use_conv_bias (`bool`, *optional*, defaults to `True`):
Whether to use bias in the convolution layer.
hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
The non-linear activation function (function or string) to use in the encoder and pooler. If string,
`"gelu"`, `"relu"`, `"swish"` and `"gelu_new"` are supported.
initializer_range (`float`, *optional*, defaults to 0.1):
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
residual_in_fp32 (`bool`, *optional*, defaults to `True`):
Whether to compute the residual connection in float32.
time_step_rank (`str` or `int`, *optional*, defaults to `"auto"`):
The rank of the time step embedding. If set to `"auto"`, the rank is calculated as
`math.ceil(self.hidden_size / 16)`.
time_step_min (`float`, *optional*, defaults to 0.001):
The minimum value for the time step embedding.
time_step_max (`float`, *optional*, defaults to 0.1):
The maximum value for the time step embedding.
time_step_floor (`float`, *optional*, defaults to 1e-4):
The floor value for the time step embedding.
time_step_limit (`tuple`, *optional*, defaults to (0.0, float("inf"))):
The minimum and maximum limits for the time step.
rescale_prenorm_residual (`bool`, *optional*, defaults to `False`):
Whether to rescale the pre-norm residual.
use_cache (`bool`, *optional*, defaults to `True`):
Whether or not the model should return the last key/values attentions (not used by all models). Only
relevant if `config.is_decoder=True`.
norm_before_gate (`bool`, *optional*, defaults to `True`):
Whether to apply normalization before the gate activation.
rms_norm (`bool`, *optional*, defaults to `True`):
Whether to use root mean square normalization.
chunk_size (`int`, *optional*, defaults to 256):
Size of chunks for processing long sequences.
tie_word_embeddings (`bool`, *optional*, defaults to `False`):
Whether to tie the word embedding weights with the output projection weights.
gradient_checkpointing (`str`, *optional*, defaults to `"nothing_saveable"`):
The gradient checkpointing configuration.
"""
model_type: str = "mamba2"
def __init__(
self,
num_heads=128,
head_dim=64,
vocab_size=32768,
hidden_size=4096,
state_size=128,
num_hidden_layers=64,
layer_norm_epsilon=1e-5,
pad_token_id=1,
bos_token_id=0,
eos_token_id=2,
expand=2,
conv_kernel=4,
n_groups=8,
use_bias=False,
use_conv_bias=True,
hidden_act="silu",
initializer_range=0.1,
residual_in_fp32=True,
time_step_rank="auto",
time_step_min=0.001,
time_step_max=0.1,
time_step_floor=1e-4,
time_step_limit=(0.0, float("inf")),
rescale_prenorm_residual=False,
use_cache=True,
norm_before_gate=True,
rms_norm=True,
chunk_size=256,
tie_word_embeddings=False,
gradient_checkpointing: EasyDeLGradientCheckPointers = EasyDeLGradientCheckPointers.NONE,
**kwargs,
):
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.state_size = state_size
self.num_hidden_layers = num_hidden_layers
self.layer_norm_epsilon = layer_norm_epsilon
self.conv_kernel = conv_kernel
self.expand = expand
self.bos_token_id = bos_token_id
self.eos_token_id = eos_token_id
self.pad_token_id = pad_token_id
self.use_bias = use_bias
self.use_conv_bias = use_conv_bias
self.hidden_act = hidden_act
self.initializer_range = initializer_range
self.time_step_rank = math.ceil(self.hidden_size / 16) if time_step_rank == "auto" else time_step_rank
self.time_step_min = time_step_min
self.time_step_max = time_step_max
self.time_step_floor = time_step_floor
self.rescale_prenorm_residual = rescale_prenorm_residual
self.residual_in_fp32 = residual_in_fp32
self.use_cache = use_cache
self.n_groups = n_groups
self.num_heads = num_heads
self.head_dim = head_dim
self.norm_before_gate = norm_before_gate
self.rms_norm = rms_norm
self.state_size = state_size
self.chunk_size = chunk_size
self.time_step_limit = time_step_limit
self.tie_word_embeddings = tie_word_embeddings
self.gradient_checkpointing = gradient_checkpointing
self.intermediate_size = int(expand * hidden_size)
super().__init__(**kwargs)
[docs] def get_partition_rules(self, *args, **kwargs):
"""
Get the partition rules for distributing the Mamba2 model parameters across multiple devices.
These rules define how parameters should be partitioned when using techniques like
Fully Sharded Data Parallelism (FSDP), Sharded Parallelism (SP), and Tensor Parallelism (TP).
Each rule consists of a regex pattern matching parameter names and a corresponding PartitionSpec.
Returns:
tuple: A tuple of tuples where each inner tuple contains:
- A regex pattern matching parameter names
- A PartitionSpec object specifying how to partition matching parameters
"""
return (
# Embeddings
("backbone/embeddings/embedding", PartitionSpec(("fsdp", "sp"), "tp")),
# Language model head
("lm_head/kernel", PartitionSpec(("fsdp", "sp"), "tp")),
("lm_head/bias", PartitionSpec("tp")),
# Mixer layers
("mixer/in_proj/kernel", PartitionSpec(("fsdp", "sp"), "tp")),
("mixer/in_proj/bias", PartitionSpec("tp")),
("mixer/out_proj/kernel", PartitionSpec("tp", ("fsdp", "sp"))),
("mixer/out_proj/bias", PartitionSpec(("fsdp", "sp"))),
# Conv1d in mixer (3D kernel)
("mixer/conv1d/kernel", PartitionSpec(None, None, "tp")),
("mixer/conv1d/bias", PartitionSpec("tp")),
# State space parameters
("mixer/A_log", PartitionSpec(None)),
("mixer/D", PartitionSpec(None)),
("mixer/dt_bias", PartitionSpec(None)),
# Normalization layers
("mixer/norm/kernel", PartitionSpec(None)),
("backbone/layers/.*/norm/kernel", PartitionSpec(None)),
("backbone/norm_f/kernel", PartitionSpec(None)),
# Catch-all
(".*", PartitionSpec(None)),
)